LePetitPrince

Given a sequence of stimuli, fMRI data from subjects exposed to these sequence one or several models making quantitative predictions from each stimulus in the sequence, the code allows you build a flexible analysis pipeline combining functions that allow you to generate R² or r maps. Function types can be for example:

• Data compression methods (already coded or that you can add)
• Data transformation methods (standardization, convolution with a kernel,or whatever your heart desires...)
• Splitting strategies
• Encoding models
• Any task that you might find useful (and that you are willing to code)

For example, the pipeline programmed in main.py fits stimuli-representations of several lanuage models to fMRI data obtained from participants listening to an audiobook (in 9 runs). R² are computed from a nested-cross validated Ridge-regression. The pipeline runs for tuples (1 subject, 1 model), to facilitate distributing the code on a cluster.